THE RUNDOWN
Friday, 21 August 2026August 21: AI, Robotaxis & Safety Battles
Nevada greenlights 8,000 robotaxis, ChatGPT starts texting for you, and Roblox is still failing kids. Here's what matters today.
Nevada just handed Tesla, Uber, and Waymo permission to put up to 8,000 robotaxis on its roads over the next twelve months β and that single number tells you everything about how fast autonomous vehicles are moving from pilot project to actual industry. It's the kind of approval that would've sounded like science fiction five years ago. Today it's a permit application. Here's the rest of what's shaping your Thursday.
Nevada says yes to 8,000 robotaxis
Tesla, Uber, and Waymo have all received operating permits in Nevada, collectively covering up to 8,000 autonomous vehicles deploying within a year. It's the largest coordinated robotaxi greenlight in US history, and it puts serious pressure on other states to either match Nevada's regulatory appetite or fall behind in the race to define how self-driving transport actually works at scale. For Africa watching from the sidelines, this is the template being built right now.
ChatGPT will now text on your behalf
OpenAI has rolled out an Apple Messages integration that lets ChatGPT compose and send texts for you β you describe what you want to say, and the AI handles the rest. It sounds convenient until you think about it for two seconds and wonder who's actually in the conversation. This sits right at the heart of the AI assistant debate that's been brewing all year: where does the tool end and the person begin?
Google Discover gets an AI-tuned feed
Google is about to let you describe your Discover feed in plain language β tell it what you want to see, and an AI chatbot will tune the feed accordingly. It's a meaningful shift from algorithmic black box to something that at least pretends to be a conversation. Rolling out to the Google app in the coming days, it's worth watching whether this actually improves content quality or just optimises for more engagement.
OpenAI gaining ground on Anthropic with businesses
New data suggests OpenAI is closing the gap with Anthropic among enterprise users, but the more interesting detail is how willing businesses are to switch between AI providers as new models drop. That kind of loyalty-free adoption should worry investors on both sides β it means neither company has actually locked anyone in. The AI tools market is still very much up for grabs.
Roblox is still failing children, Australia says
Australia's eSafety regulator has come back to Roblox with a firm verdict: the platform hasn't done enough to protect children from predatory adults, and it's now the first platform compelled to submit to independent audits under the Online Safety Act. Meanwhile, US senators are separately pressing TikTok over a deeply troubling internal experiment β the platform reportedly withheld an algorithm safety feature from a test group that included a teen who later died by suicide. Two platforms, two countries, one very loud message about who tech companies are actually protecting.
FCC officially lowers the broadband bar
The US Federal Communications Commission, under Chairman Brendan Carr, has formally backed away from gigabit broadband as a national speed benchmark. In plain terms: the regulator responsible for America's internet ambitions has decided fast internet is a nice-to-have, not a standard. For a country trying to compete globally in AI and cloud infrastructure, that's a strange hill to plant a flag on.
A fake crypto conference was actually a malware trap
Security researchers are reporting that a hacker posing as a representative of a major crypto news site lured cybersecurity professionals into a fake conference, using Google Docs to deliver malware. Targeting the people whose job it is to catch exactly this kind of attack takes a certain audacity β and it's a reminder that social engineering doesn't need to be technically sophisticated to be effective. It just needs to be believable.
Framework's BIOS update is bricking laptops
Framework, the company that built its reputation on repairability and user control, is dealing with a painful irony: a recent BIOS update has been bricking some Framework Laptop 13 machines running last-gen AMD chips, on both Windows and Linux. The company says it's addressing the issue, but for a brand whose whole pitch is that you're in control of your own hardware, this one stings.
AI training data startup hits $500M run rate
Micro1, a startup that supplies training data for AI models, has reached a $500 million gross run rate β a figure that reflects just how hungry the big AI labs are for quality data right now. It's easy to focus on the flashy models, but the real infrastructure boom is happening at this layer: the unglamorous, essential work of feeding those models something worth learning from.
Apple Music to label AI-generated tracks
Apple Music is reportedly preparing to add transparency labels to tracks made entirely by AI β expanding its existing tagging system to flag what critics are already calling 'slop songs.' It won't stop AI music from being made, but it at least gives listeners the information to decide what they're actually listening to. Whether other streaming platforms follow matters a great deal for artists everywhere, including those building careers across West Africa's increasingly digital music scene.
That's your Thursday. The robotaxi numbers in Nevada are the story to bookmark β we're watching a regulatory template form in real time, and it will travel.
Comments
Get Weekly Tech Tips
Join 10,000+ readers getting expert tech insights delivered to their inbox.
No spam. Unsubscribe anytime.